Behaviour annotations

When a knockout changes behaviour, is behaviour the gene's function?

AI Gene Review · projects/BEHAVIOR · 2026

Bottom line

  • Behaviour terms (GO:0007610 and children) come mostly from knockout phenotypes (IMP, IGI), whatever the gene's molecular job.
  • We mined the whole corpus, wrote a four-step rubric, and mapped 16 standard behavioural assays to the GO terms they can support.
  • 86% of adjudicated behaviour rows (169 of 197) are downgraded; the 28 accepted are mostly sensory channels and receptors, plus CRY and GCG.

Why behaviour is the hard case

  • A behaviour integrates the whole nervous system, plus development, metabolism and basic cell biology.
  • So almost any perturbation can move it: a tubulin, a lysosomal peptidase, a ciliary scaffold.
  • In the ASSAY_TO_FUNCTION framing it is the most distal, most convergent readout.
  • Source surface today: 209 behaviour annotations on 87 genes in GOA files; IMP 84, IEA 39, IGI 30, IDA 3.

The chain and the rubric

What reviewers decided

Assays as evidence

Status and next steps

  • ✅ Corpus mined; rubric written; accepted rows spot-checked (9 missed downgrades fixed).
  • ✅ IMPReSS ingested; assay→GO map and checker built; BEHAVIORAL_ASSAY class added to ASSAY_TO_FUNCTION.
  • ✅ reports/REPORT.md regenerated (September 2026: 197 adjudicated, 86%).
  • ⬜ Resolve 9 UNDECIDED rows, including Agtr1a drinking (IMP/IGI).
  • ⬜ Record which assay drove each behaviour annotation so the check can run automatically.

Read more: projects/BEHAVIOR.md · projects/BEHAVIOR/impress/ · projects/BEHAVIOR/mine_behavior.py